# MTH3230 Exam Sheet .pdf

### File information

Original filename: MTH3230 Exam Sheet.pdf

This PDF 1.5 document has been generated by TeX / MiKTeX pdfTeX-1.40.17, and has been sent on pdf-archive.com on 14/12/2017 at 17:46, from IP address 110.22.x.x. The current document download page has been viewed 393 times.
File size: 210 KB (2 pages).
Privacy: public file

MTH3230 Exam Sheet.pdf (PDF, 210 KB)

### Document preview

Sequence Definitions
δ = (. . . , 0, 1 , 0 . . . ), U = (. . . , 0, 1 , 1 . . . )
U0

δ0

Periodic Inputs =⇒ ∃(a ∈ CN ) :
N
−1
X
2πkn
ak ei N
xn =

Identity System
(I ◦ x)n = xn
Backshift Operator
(B ◦ x)n = xn−1

ARMA(p, q)
p
q
X
X
θk Zn−k
Xn −
φj Xn−j = Zn +
j=1

k=0

yn =

N
−1
X

i 2πk
N

H(e

)ak e

Weak Stationarity
E[Xn ] = constant ∀(n ∈ Z)
E[Xn Xn+h ] = γ(h) ∀(n, ∈ Z)

k=0

xn ∈ R =⇒ X(eiλ ) = X(eiλ )

Convolution
X
xk x
ˆn−k
(x ∗ x
ˆ)n =

H(eiλ ) =

k∈Z

Impulse Response
hn = (T ◦ δ)n
yn = (T ◦ x)n = (x ∗ h)n

b0 + b1 e−iλ + · · · + bp e−iλp
1 + a1 e−iλ + · · · + aq e−iλq

Linearity
T ◦ (ax + bˆ
x) = aT ◦ x + bT ◦ x
ˆ

Orthogonality. E[XY ] = 0
Projection
Yˆ = a1 X1 + · · · + an Xn
E[(Y − Yˆ )Yi ] = 0

Fundamental Theorem of Algebra
k
X
p(z) =
cj z j , cj ∈ R

Mean-Square-Error
M SE = E[(Y − Yˆ )2 ]

j=0

Stability

Partial Fraction Decomposition
1
A1
Ak
=
+ ··· +
p(z)
1 − a1 z −1
1 − ak z −1

z q /Y (z) = 0 =⇒ |z| &lt; 1
X
|hn | &lt; ∞
or

Difference of Squares
a2 − b2 = (a + b)(a − b)

n∈Z

ˆ
Uniqueness. If ∀(z ∈ ROC), X(z) = X(z),
then xn = x
ˆn
Convolution Theorem yn = (x ∗ h)n
Transfer Function
X
H(z) =
hn z −n = Y (z)/X(z)

H(z) =

n∈N

z −p

P (z)
b0 + b1
+ · · · + bp
=
1 + a1 z −1 + · · · + aq z −q
Q(z)

Frequency Response = H(eiλ )
N
X
If xn =
ak eirk n
k=0

Then yn =

N
X

H(eirk )ak eirk n

k=0

Sequence
xn

Autocovariance Xn =

Step Function

δn
Un
an Un
n+k−1 n
a Un
k−1

Trigonometric

Linearity

g(x)fX (x) dx

h∈Z

f (λ) = f (λ + 2nπ), f (λ) = f (−λ)
Inverse ZFourier TransformZ
π
π
γ(h) =
eihλ f (λ) dλ =
cos(hλ)f (λ) dλ
−π
−π
X
General ARMA Yn =
ψk Xn−k

Variance
var(X) = E[X 2 ] − E[X]2
Covariance
cov(X) = E[XY ] − E[X]E[Y ]
Correlation
cov(X, Y )
ρ(X, Y ) = p
var(X) var(Y )
Autocovariance
γ(m, n) = cov(Xn , Xm )
γ(h) = γ(−h)

k∈Z

γY (h) =

ψk ψj γX (h + k − j)

k∈Z j∈Z

Z–Transform

Discrete F –Transform

X

X

xn z −n

XX

fY (λ) = |Ψ(e−iλ )|2 fX (λ)

X(eiλ ) =

xn e−iλn

Inversion
I
1
xn = 2πi
X(z)z n−1 dz
C

n∈Z

1

1

1
1−z −1

1
1−e−iλ

1
1−az −1

1
1−ae−iλ

k
1
1−az −1

k
1
1−ae−iλ

cos(ωn)Un

1−z −1 cos ω
(z −1 −cos ω)2 +sin2 ω

1−e−iλ cos ω
(e−iλ −cos ω)2 +sin2 ω

sin(ωn)Un

z −1 sin ω
(z −1 −cos ω)2 +sin2 ω

e−iλ sin ω
(e−iλ −cos ω)2 +sin2 ω

axn + bˆ
xn

ˆ
aX(z) + bX(z)

ˆ iλ )
aX(eiλ ) + bX(e

xn−k

z −k X(z)

e−ikλ X(eiλ )

Modulation

an xn

X( az )

X( ea )

nxn

d
−z dz
X(z)

d
i dλ
X(eiλ )

(x ∗ x
ˆ)n

ˆ
X(z)X(z)

(x · x
ˆ)n

ψk Zn−k

k∈Z

Shifting

Convolution

X

Spectral Density ∈ R &gt; 0
1
1 X
f (λ) =
Γ(eiλ ) =
γ(h)e−ihλ

2π h∈Z
X
1
= 2π
γ(h) cos(λh)

Cauchy-Schwarz Inequality
E[XY ]2 6 E[X 2 ]E[Y 2 ]

X(z) =

j = 1, . . . , n

k∈Z
X
ψk ψk+h
γ(h) = σ 2

x∈Z
Z

n∈Z

Delta Function

ak E[Xk Xj ] = E[Y Xj ]

k=0

R

k=0

z −1

n
X

Expectation
X
E[g(X)] =
g(x)P(X = x)
E[g(X)] =

j = 0, . . . , n

k=0

Euler’s Formulæ, eix = cos x + i sin x
1
cos x = 21 (eix +e−ix ), sin x = 2i
(eix −e−ix )

n∈Z

j=1

Yule-Walker Equations
n
X
ak Xk )Xj ] = 0
E[(Y −

∃λ ∈ Ck : p(z) = d(z −λ1 )(z −λ2 ) . . . (z −λk )

Causality
No dependence on future values,
or ∀(n &lt; 0), hn = 0 ∨ xn = 0

General Difference Equation
q
p
X
X
yn = −
aj yn−j +
bk xn−k ,

Causality. ∀(|z| &lt; 1), Φ(z) 6= 0
Invertibility. ∀(|z| &lt; 1), Θ(z) 6= 0

Geometric Series (|a| &lt; 1)
X
1
an =
1

a
n∈N

Time Invariance
T ◦B◦x=B◦T ◦x

k=1

Φ(B)Xn = Θ(B)Zn

i 2πkn
N

ˆ iλ )
X(eiλ )X(e
1

Z

π
−π

ˆ i(λ−s) ) ds
X(eis )X(e

xn =

1

π

Z

X(eiλ )eiλn dλ

−π

If Xn = Ψ(B)Zn
f (λ) =

σ2
|Ψ(e−iλ )|2

Ideal Low-Pass Filter (
H(eiλ ) = 1{|λ|&lt;C} =
(
hn =

C
π

sinc( C
n)
π

=

= 1{|λ|&gt;C} =

(
1 − C/π
hn =
− sin(Cn)/πn

Statistically Designed Filters
Wn ∼ N (0, ν 2 )

if |λ| &lt; C
if |λ| &gt; C

C/π
sin(Cn)/πn

Ideal High-Pass Filter (
H(eiλ )

1
0

if n = 0
if n =
6 0

Wiener Filter Y = Xn + Wn
fXY (λ) ?
fX (λ)
A(eiλ ) =
=
fY (λ)
fX (λ) + ν 2 /2π
Mean-Square-Error

0
1

if |λ| &lt; C
if |λ| &gt; C

ˆ n )2 = γX (0) −
E(Xn − X

N
X

aj γXY (j)

j=0

if n = 0
if n =
6 0

π

Z

fX (λ) −

=
−π

|fXY (λ)|2

fY (λ)

H(eiλ ) + H(eiλ ) = 1{|λ|&lt;C} + 1{|λ|&gt;C} = 1 Adaptive Projection
Ln = hYn , Yn−1 , . . . , Y0 i
=⇒ hn + hn = δn
P (X|Ln+1 ) = P (X|Ln )
Statistical Mean
+ β(Yn+1 − P (Yn+1 |Ln ))
x1 + · · · + xN
where
x=
E[(X−P (X|Ln ))(Yn+1 −P (Yn+1 |Ln ))]
β=
N
E[(Y
−P (Y
|L ))2 ]
n+1

Sample Autocovariance
N −|h|
1 X
γ
ˆ (h) =
(xn − x)(xn+|h| − x)
N n=1
Sample Autocorrelation
ρˆ(h) =

γ
ˆ (h)
γ
ˆ (0)

n

L2 Norm, Inner Product
hX, Y i = E[XY ] = E[Y X] = hY, Xi
haX, Y i = aE[XY ] = hX, aY i
hX1 + Xp
2 , Y i = hXp
1 , Y i + hX2 , Y i
||X|| = E[X 2 ] = hX, Xi
||X||2 = E[X 2 ] = hX, Xi
|hX, Y i| = |E[XY ]| 6 ||X|| × ||Y ||
||X + Y || 6 ||X|| + ||Y || (Triangle Inequality)
||X + Y ||2 = ||X||2 + ||Y ||2 when E[XY ] = 0
Projection Theorem
∃Yˆ (unique)
||Y − Yˆ || 6 ||Y − X|| ∀X ∈ K
If X ∈ K ∧ ||Y − Yˆ || = ||Y − X||
Then X = Yˆ
P (aY + bZ | K) = aP (X | K) + bP (Z | K)

Kalman Filter Yn = Xn + Wn
P (Xn+1 |Ln+1 ) = (1 − βn )P (Xn+1 |Ln )
+ βn Yn+1
where
E[(Xn+1 − P (Xn+1 |Ln ))2 ]
βn =
E[(Xn+1 − P (Xn+1 |Ln ))2 ] + ν 2
and

Periodogram
IN (λ) =

n+1

Mean-Square-Error
E(X − P (X|Ln+1 ))2 = E[(X − P (X|Ln ))]2
E[(X − P (X|Ln ))(Yn+1 − P (Yn+1 |Ln ))]2

E[(Yn+1 − P (Yn+1 |Ln ))2 ]

R2 Norm
v · w = v1 w1 + v2 w2 = |v||w| cos θ = w · v
(av) · w = v · (aw) = a(v · w)
(u + v) · w = u · w + v · w
|v|2 = v12 + v22 = v · v
|v · w| = |v||w|| cos θ| 6 |v||w|
|v + w| 6 |v| + |w| (Triangle Inequality)
|v + w|2 = |v|2 + |w|2 when v · w = 0

N
2
1 X

xk e−iλk

N k=1

Vn+1 = E[(Xn+1 − P (Xn+1 |Ln+1 ))2 ]
=

E[(Xn+1 − P (Xn+1 |Ln ))2 ]ν 2
E[(Xn+1 − P (Xn+1 |Ln ))2 ] + ν 2

Kalman Gain

V∞ is determined by the unique positive
root of
(a2 V∞ + σ 2 )ν 2
V∞ = 2
a V∞ + σ 2 + ν 2

P (Xn+1 |Ln+1 ) =

ν2
V∞ is determined by the unique positive
P
(X
|L
)
n
n+1
E[(Xn+1 − P (Xn+1 |Ln ))2 ] + ν 2
root of
E[(Xn+1 − P (Xn+1 |Ln ))2 ]
(a2 ν 2 + σ 2 )V∞ + ν 2 σ 2
Yn+1
+
V∞ =
2
2
E[(Xn+1 − P (Xn+1 |Ln )) ] + ν
V∞ + ν 2
Process: Xn = aXn−1 + Zn , Observation Proces: Yn = Xn + Wn
Kalman Equations

P (Xn+1 |Ln+1 ) =

E[(X
2
ν2
n+1 − P (Xn+1 |Ln )) ]
P (Xn+1 |Ln ) +
Yn+1
2
2
2
2
E[(Xn+1 − P (Xn+1 |Ln )) ] + ν
E[(Xn+1 − P (Xn+1 |Ln )) ] + ν

Vn+1 = E[(Xn+1 − P (Xn+1 |Ln+1 ))2 ] =

E[(Xn+1 − P (Xn+1 |Ln ))2 ]ν 2
E[(Xn+1 − P (Xn+1 |Ln ))2 ] + ν 2

Kalman Estimation

a2 V + σ 2

ν2
n
ˆ n+1 =
ˆn +
aX
Yn+1
X
2
2
2
a Vn + σ + ν
a2 Vn + σ 2 + ν 2
Vn+1 =

(a2 Vn + σ 2 )ν 2
a2 Vn + σ 2 + ν 2

ˆ 0 = P (X0 |Y0 ) and V0 = E[(X0 − X
ˆ 0 )2 ]
with X

Kalman Predication
V

ν2
n
ˆ n+1 =
ˆn +
aX
aYn
X
Vn + ν 2
Vn + ν 2
Vn+1 =

(a2 ν 2 + σ 2 )Vn + ν 2 σ 2
Vn + ν 2

ˆ 1 = P (X1 |Y0 ) and V1 = E[(X1 − X
ˆ 1 )2 ]
with X